CN115878986A - Degradation trend evaluation method for variable working condition bearing - Google Patents
Degradation trend evaluation method for variable working condition bearing Download PDFInfo
- Publication number
- CN115878986A CN115878986A CN202211493210.XA CN202211493210A CN115878986A CN 115878986 A CN115878986 A CN 115878986A CN 202211493210 A CN202211493210 A CN 202211493210A CN 115878986 A CN115878986 A CN 115878986A
- Authority
- CN
- China
- Prior art keywords
- bearing
- module
- characteristic
- modal
- degradation trend
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a degradation trend evaluation method of a variable working condition bearing, and belongs to the technical field of equipment state monitoring. According to the method, the probability distribution of the random convolution kernel is initialized to the probability distribution of the bearing data, so that the decoupling of the bearing signal containing noise coupling into the multi-scale characteristic representation of the bearing fault related characteristic and the bearing fault independent characteristic by the random convolution kernel is facilitated, the model generalization of the condition (namely an unknown domain) without the bearing is promoted, and the model can accurately judge the degradation trend under the bearing signals of different conditions; because the random convolution kernel transformation is only equivalent to a single-layer network, the parameter quantity is 1-2 orders of magnitude lower than that of a feature extraction network in the prior art, and the method has a remarkable speed advantage; the health state characteristics of the bearing vibration signals are constructed by the modal parameters and the time domain frequency domain parameters, and the generalization and accuracy of the model are further improved.
Description
Technical Field
The invention belongs to the technical field of equipment state monitoring, and particularly relates to a degradation trend evaluation method of a variable working condition bearing.
Background
A bearing is a most commonly used core-mounted mechanical component in a rotary machine, and is a high-frequency component having a failure due to the dispersion of life. When the bearing operates in a normal working environment, the bearing is influenced by a plurality of objective conditions such as load, rotating speed, temperature, lubrication and the like, and the failure of different positions, different degrees and different reasons can be caused inevitably, such as the abrasion, the gluing and the fracture of an inner ring, an outer ring, a roller and a retainer. Whether the bearing works normally or not is a core problem of state monitoring of the rotating machinery as the most basic and widely applied mechanical part in a rotating structure. As such, the prediction of the degradation trend of the rolling bearing and the health status evaluation have a significant influence on the whole mechanical industry.
The health condition characterization parameters of the evaluation method of the bearing degradation trend are all from vibration signals in the bearing working process, and the feature extraction method of the vibration signals is divided into two types according to feature dimensions: only one-dimensional and multi-dimensional feature vectors are used. The one-dimensional characteristic vector has a series of problems of poor interference, less contained information amount and the like, and the health condition and the degradation trend of the bearing in the whole life cycle are difficult to accurately and effectively evaluate; the method using the multi-dimensional feature vector has stronger anti-jamming capability and more comprehensive feature expression, and the multi-dimensional vector input has higher prediction precision than a one-dimensional feature vector input model, but the calculation complexity is higher than the one-dimensional feature input model. The existing bearing degradation trend machine learning model is poor in generalization, a deep learning model is large in sample requirement, long in model training and reasoning time and insufficient in calculation capacity of an equipment edge end, and the current situation of a lightweight and high-accuracy online trend evaluation model is urgently needed, so that the existing technical scheme is difficult to realize industrial online bearing degradation trend evaluation.
Disclosure of Invention
Aiming at the defects or improvement requirements of the prior art, the invention provides a degradation trend evaluation method of a variable working condition bearing, and aims to improve the generalization of a machine learning model of the degradation trend of the bearing and improve the evaluation accuracy.
To achieve the above object, according to one aspect of the present invention, there is provided a degradation tendency evaluation method of a variable condition bearing, comprising:
s1, constructing a bearing degradation trend evaluation model; the bearing degradation trend evaluation model comprises a feature extraction module, a feature fusion module, a random convolution kernel transformation module, a dimension reduction module and a label classifier; the characteristic extraction module is used for extracting modal characteristics and time-frequency characteristics of the input data; the characteristic fusion module is used for carrying out characteristic fusion on the time-frequency characteristic and the modal characteristic through typical correlation analysis; the random convolution kernel transformation module is used for performing convolution operation on the fused characteristic parameters through a large number of multi-scale convolution kernels and extracting multi-scale characteristics related to bearing faults and unrelated to the bearing faults; the dimension reduction module is used for carrying out dimension reduction projection on the multi-scale features and constructing health state features; the label classifier is used for mapping the health state features subjected to dimension reduction into corresponding predicted bearing health states;
s2, using vibration signal data of the existing multi-working-condition bearing with the label as source domain data, and using vibration signal data of the bearing without the label in the new working condition as target domain data to form a training sample set;
s3, carrying out iterative training on the bearing degradation trend evaluation model by adopting a training sample set; in the training process, initializing the joint probability distribution of the random convolution kernels to be consistent with the data distribution of a bearing training set; the length of the random convolution kernel is set to be the product of the number value of sampling points of more than one circle of the rolling bearing and the normal distribution
And S4, inputting vibration data of the bearing to be tested under the current working condition into the trained model to obtain a bearing degradation trend evaluation result.
Further, the fusion process of the modal characteristics and the time-frequency characteristics is as follows:
analyzing and fusing time-frequency characteristics through typical correlation;
and multiplying the structural modal characteristics by the corresponding coefficients, and performing concat secondary fusion on the time-frequency characteristics subjected to typical correlation analysis.
Further, acquiring modal characteristics by using a covariance-based random subspace identification method SSICOV; the modal characteristics include modal frequency, modal shape and modal damping.
Further, the modal characteristics are also preprocessed as follows: taking the first-order natural frequency from the multi-order natural frequency parameters as an effective value; and meanwhile, interpolation is adopted to process missing values.
Further, the time domain features are root mean square values and kurtosis indexes.
Further, the frequency domain characteristic is a root mean square frequency.
The invention also provides a degradation trend evaluation device of the variable working condition bearing, which comprises the following components:
the evaluation model building module is used for building a bearing degradation trend evaluation model; the bearing degradation trend evaluation model comprises a feature extraction module, a feature fusion module, a random convolution kernel transformation module, a dimension reduction module and a label classifier; the characteristic extraction module is used for extracting modal characteristics and time-frequency characteristics of the input data; the characteristic fusion module is used for carrying out characteristic fusion on the time-frequency characteristic and the modal characteristic through typical correlation analysis; the random convolution kernel transformation module is used for performing convolution operation on the fused characteristic parameters through a large number of multi-scale convolution kernels and extracting multi-scale characteristics related to bearing faults and irrelevant to the bearing faults; the dimension reduction module is used for carrying out dimension reduction projection on the multi-scale features and constructing health state features; the label classifier is used for mapping the health state features subjected to dimension reduction into corresponding predicted bearing health states;
the training sample construction module is used for taking vibration signal data of an existing multi-working-condition bearing with a label as source domain data and taking vibration signal data of a bearing without a label under a new working condition as target domain data to form a training sample set;
the iterative training module is used for performing iterative training on the bearing degradation trend evaluation model by adopting a training sample set; in the training process, initializing the joint probability distribution of the random convolution kernels to be consistent with the data distribution of a bearing training set; the length of the random convolution kernel is set to be the product of the number value of sampling points of more than one circle of the rolling bearing and the normal distribution
And the online evaluation module is used for inputting the vibration data of the bearing to be tested under the current working condition into the trained model to obtain the evaluation result of the degradation trend of the bearing.
In general, the above technical solutions conceived by the present invention can achieve the following advantageous effects compared to the prior art.
(1) Different from the random initialization of the traditional random convolution kernel, the probability distribution of the random convolution kernel is initialized into the probability distribution of the bearing data, which is beneficial to decoupling the bearing signal containing noise coupling by the random convolution kernel into the multi-scale characteristic representation of the bearing fault related characteristic and the bearing fault independent characteristic, thereby improving the model generalization of the working condition (namely unknown domain) of the bearing which is not seen, and realizing that the bearing signal model can also accurately judge the degradation trend under different working conditions; meanwhile, the random convolution kernel transformation is only equivalent to a single-layer network, and the parameter quantity is 1-2 orders of magnitude lower than that of a feature extraction network, so that the method has a remarkable speed advantage.
(2) The method adopts the modal parameters and the time domain frequency domain parameters to jointly construct the health state characteristics of the bearing vibration signals. Because the dynamic characteristics of the system structure are changed along with the degradation of the mechanical system, the natural frequency and the damping ratio of the system are changed along with the degradation of the system, the modal characteristics represented by the natural frequency and the damping ratio can be used as the characteristics with prior information for representing the degradation trend of the bearing, and the solution space of the model can be effectively reduced through the integration of the prior characteristics, so that the model focuses more on the generalizable characteristics related to the degradation of the bearing, and the generalization and the accuracy of the model are improved.
Drawings
FIG. 1 is a flowchart of a method for constructing a degradation trend evaluation model of a variable condition bearing according to embodiment 1 of the present invention;
FIG. 2 is a schematic structural diagram of a degradation trend evaluation model of a variable condition bearing according to embodiment 1 of the present invention;
fig. 3 is a schematic diagram of a logic for performing time series data prediction by using the random convolution kernel transform model according to embodiment 1 of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the respective embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Example 1
A degradation trend evaluation method for a variable working condition bearing takes data of a bearing data set of the Siann transport university as input, then feature extraction is carried out, a plurality of other linearly independent feature quantities (including time domain, frequency domain and modal feature quantities) are used for mechanical learning and derivation of selected health index parameters, then data training is carried out based on random convolution kernel transformation, and therefore an accurate label classifier is obtained, and a degradation trend evaluation model of the variable working condition bearing is further established.
Specifically, as shown in fig. 1, the method comprises the following steps:
s1, building a bearing degradation trend evaluation model;
referring to fig. 2, the bearing degradation tendency evaluation model includes: the system comprises a feature extraction module, a feature fusion module, a random convolution kernel transformation module, a dimension reduction module and a label classifier;
the characteristic extraction module is used for extracting modal and time-frequency characteristics of input data; the characteristic fusion module is used for carrying out characteristic fusion on the time-frequency characteristic and the modal characteristic through typical correlation analysis; the random convolution kernel transformation module is used for processing the fused characteristic parameters to obtain multi-scale characteristics; the dimension reduction module is used for carrying out dimension reduction projection on the multi-scale features and constructing health state features; the label classifier is used for mapping the health state features after dimension reduction into corresponding predicted bearing health states;
the establishment of the health status indicator will be described in several ways below:
obtaining modal information: the modal parameters typically include modal frequency, modal damping, and modal shape.
As the vibration signals generate covariance shift along with the degradation of the system, and the Toeplitz matrix constructed by the covariance-based random subspace identification method SSI-COV reflects the change of the covariance, the modal frequency, the modal shape and the modal damping representing the structural state of the current equipment are obtained through the SSICOV function. And processing a complete group of files (full life cycle vibration signals of a certain primary bearing in a data set) according to the same SSICOV modal parameter identification method to obtain a full life cycle modal frequency, modal damping and a modal vibration mode health state index parameter variation trend matrix.
It should be noted that the feature extraction module also preprocesses the mode parameters, for example, most of the natural frequency parameters obtained by the SSICOV function are more than one, and there may be no corresponding natural frequency at a certain time. Therefore, the first-order natural frequency is taken as a basic parameter used by a model experiment only in the excited multi-order natural frequency parameters, for example, the modal frequency only selects the first data of each item as a sample, and under the condition that a missing value exists, the missing value is considered to be processed by interpolation, and modal damping and modal shape health state indexes are converted into column vectors through a similar method to be used as a part of input of the feature fusion module.
Acquiring time domain information: in the time domain characteristics of the bearing vibration data, the emphasis points of reactions of dimensional parameters and dimensionless parameters are different: the dimensional characteristic indexes are mostly related to objective working environments such as load, rotating speed and the like, and the dimensionless characteristic tends to intuitively display impact information of the operation process of the bearing component. If classified according to the characteristics of the bearing vibration data, the early-stage degradation damage of the bearing is roughly classified into two types: surface damage and wear. Dimensionless parameters can effectively identify the former, but are not sensitive to the latter; the opposite is true with dimensional parameters, which are very sensitive to the latter but have little response to the former. In addition, the time domain characteristic parameters of the bearing vibration data are slightly different from the emphasis points of the bearing health condition description. The root mean square can effectively reflect the vibration condition of the bearing component, describes the variation trend of the amplitude and the energy of the vibration signal in the time domain, and is therefore more suitable for representing the current health condition of the bearing. The kurtosis index and the skewness index can reflect the difference between the vibration signal of the bearing and the normal distribution at the current moment (if the bearing is in a normal operation condition, the amplitude of the vibration signal of the bearing is approximately consistent with the normal distribution), which indicates that the bearing can sensitively find the surface damage type fault of the bearing. Generally, any time domain feature data cannot reflect the degradation trend of the bearing in the whole life cycle more completely, so that dimensional parameters and dimensionless parameters are respectively extracted and taken as references. In this embodiment, the root mean square value and kurtosis index are taken for standby.
Inputting bearing vibration acceleration data, and calculating by a formula (number) to obtain a root mean square value and a kurtosis index.
Wherein
x (N) and N indicate that each acquisition point has N data, and the vibration displacement corresponding to each data is x (N).
And processing a complete group of files (full-life-cycle vibration signals of a certain bearing in a data set) according to a similar method to obtain the parameters of the root mean square value and the kurtosis value of the experiment.
Acquiring frequency domain information: the frequency domain characteristics of the vibration data reflect the frequency components contained in the signal and their energy magnitudes. The method for describing the health condition of the bearing by using the frequency domain characteristic data mainly comprises the steps of changing frequency components, energy and main frequency positions in a certain time, wherein most of different frequency domain characteristics are mainly focused on the change of a certain item, and root mean square frequency is selected as the characteristic data adopted at this time.
The acquisition mode is basically the same as the acquisition mode of time sequence data, and the required root mean square frequency parameter is obtained after the data is subjected to Fourier transform and then formula processing.
Wherein
s (k) is a frequency spectrum signal obtained by Fourier transform of the time domain signal x (n); f. of k The frequency amplitude value of the corresponding point is obtained; n is a radical of fft Is the length of the spectral signal.
Next, feature fusion is required to be performed, and further, a characterization parameter (health status index) as a bearing health condition is constructed. As the dynamic characteristics of the system structure are changed along with the degradation of the mechanical system, the natural frequency, the damping ratio and the like of the system are changed along with the degradation of the system, modal characteristics represented by the natural frequency and the damping ratio can be used as characteristics with prior information for representing the degradation trend of the bearing, and the solution space of the model can be effectively reduced through the integration of the prior characteristics, so that the model focuses more on the generalizable characteristics related to the degradation of the bearing, and the generalization and the accuracy of the model are improved.
The direct splicing characteristic fusion can cause modal characteristics to be submerged in numerous time-frequency characteristics, and the change guidance of structural modal parameters along with system degradation cannot be embodied, so the invention provides a two-stage characteristic fusion mode, wherein the first stage characteristic correlation analysis fusion time-frequency characteristics are firstly used, and then the structural modal characteristics are multiplied by coefficients to perform concat secondary fusion with the time-frequency characteristics after the typical correlation analysis; and the modal characteristic coefficient is determined according to the bearing modal contribution of the application scene. The significance degree of the fusion of the dynamic characteristics and the time-frequency domain characteristics can be effectively improved through the two-stage characteristic fusion;
s2, using vibration signal data of an existing multi-working-condition bearing with a label as source domain data, and using vibration signal data of a bearing without a label under a new working condition as target domain data to form a training sample set;
and S3, referring to the graph 3, inputting training data into the model to train the model, and performing primary verification on the modeling effect under the same distribution test data, such as performing modeling effect test on different bearing fault data under the same working condition. After analysis, the model is tried to predict the vibration signal data of the bearing without the new working condition label, and finally, the coincidence degree of the predicted value and the actual value is compared to judge whether the model is in accordance with expectation.
The specific implementation of the method is as follows:
firstly, defining a training dimension in a random convolution kernel transformation module and fitting a classifier.
Different from the random initialization of the traditional random convolution kernel, the method initializes the joint probability distribution of the random convolution kernel so as to lead the joint probability distribution to be consistent with the data distribution of the bearing training set. The probability distribution of the random convolution kernels is initialized to the probability distribution of the bearing data, so that decoupling of noise-containing coupled bearing signals of the random convolution kernels into multi-scale characteristic representation of bearing fault related characteristics and bearing fault independent characteristics can be facilitated, model generalization of the bearing signal model under the condition that the bearing is not seen (namely an unknown domain) is improved, and the degradation trend of the bearing signal model can be accurately judged under different conditions.
In order to enable the model to focus on the periodic characteristic rule of the rolling bearing fault, the length of the random convolution kernel is set to be the product of the value C of one circle of sampling points of the rolling bearing and the normal distribution. The weights, biases, expansions, steps of the random convolution kernel are initialized with the positive-error distribution. When filling each kernel, the random convolution kernel transformation module will decide whether to use the filling when applying the kernel randomly and with equal probability.
The fused and dimensionality-reduced features are used to train a linear classifier. The linear classifiers are further classified into ridge regression classifiers and logistic regression classifiers. When the number of training samples is less than the number of features, a ridge regression classifier is used, whereas a logistic regression classifier is used.
In addition, when the random convolution kernel transformation model is combined with the classifier, a single-layer convolution neural network with random kernel weight is actually formed, the parameter quantity of the model is 1-2 orders of magnitude lower than that of a feature extraction network model in the conventional technology, and the model has a remarkable speed advantage and lower requirement on computing resources for an online application scene.
The random convolution kernel transformation process is as follows:
wherein, X i Is the ith position of the time sequence; omega is a convolution kernel; l kernel Is the length of the convolution kernel; and d is the coefficient of expansion.
After random convolution kernel transformation, two feature values are obtained: maximum (equivalent to global maximum pooling) and positive values.
And finally, inputting the health state characteristics into a label classifier for training, wherein the degradation stage of the bearing degradation trend evaluation is an integer value, and the decimal output by model prediction is required to be reduced to an integer in the training process. Considering that the bearing degradation process is divided into 5 stages, namely a normal stage, an initial failure stage, a development stage, a rapid development stage and a final stage of the bearing, the model prediction result is normalized by 0-1 and then multiplied by a dimensionless constant 5 to obtain a health state corresponding to the true bearing degradation trend.
And S4, inputting the data to be tested into the trained model after the model training is finished, and evaluating the degradation trend of the bearing under the current working condition.
Example 2
A degradation trend evaluation method for a variable condition bearing comprises the following steps:
inputting bearing vibration signal data to be diagnosed into a degradation trend evaluation model of the variable condition bearing constructed by the construction method of the degradation trend evaluation model of the variable condition bearing provided by embodiment 1 of the invention, extracting modal and time domain features based on a feature extraction module in the degradation trend evaluation model of the variable condition bearing, then performing feature fusion, constructing a health state index, and inputting the index into a label classifier in the degradation trend evaluation model of the variable condition bearing, thereby obtaining a fault category label of the bearing vibration signal data to be diagnosed.
1) Preparing a data set:
the bearing data set of the university of Xian traffic is taken. Each data set has multiple stages of bearing degradation, such as bearing failure early, developing, fast developing, and end.
Loading multi-period data and dividing a training set and a test set; and carrying out standardization processing on the data in the divided training set, and selecting partial working condition combinations as source domains, such as data of 35Hz 12kN Bearing1 _1and 37.5Hz 11kN Bearing2 _1as source domains, and data of 35Hz 12kN Bearing1 _2and 37.5Hz 111kN Bearing2 _2as target domains.
2) Training a degradation trend evaluation model of the variable-condition bearing:
the method of the embodiment 1 of the invention is adopted to train the fault diagnosis model of the bearing under variable working conditions. The random convolution kernel transform uses a convolution kernel transform time series, as in a typical convolutional neural network. Essentially, any aspect of the kernel is random, such as length, weight, bias, inflation, and fill. Through experimental testing, for each core, these values were set as follows:
the length is randomly chosen from {7,9,11} with equal probability to ensure that the convolution kernel is much shorter than the input time series in most cases; sampling the weight from a standard normal distribution, and taking the average value after setting; the deviations are sampled from a uniform distribution of (-1, 1) and are only related to the positive ratio in the feature map; dilation is sampled on a base-2 exponential scale to ensure that the effective length of the kernel (including dilation) is equal to the length of the input time series, allowing kernels that are otherwise similar, but with different dilations, to have to match the same or similar patterns at different frequencies and scales; when filling generates each kernel, whether filling is used or not is decided (randomly and with equal probability) when the kernel is applied, if filling is used, a certain amount of zero filling is added at the beginning and the end of each time sequence when the kernel is applied, so that the middle element of the kernel is centered at each point in the time sequence; the stride is then always taken to be 1.
3) Evaluating the degradation trend of the variable condition bearing:
and inputting the vibration signal data of the bearing without the label in the test set into a trained degradation trend evaluation model of the variable condition bearing, and after feature extraction, directly inputting the extracted features into a label classifier to evaluate the degradation trend of the variable condition bearing.
In summary, in the embodiment, data of a bearing data set of the university of west ampere transportation is used as input, and meanwhile, appropriate parameters are selected to characterize and verify the health condition of the bearing. And establishing a degradation trend evaluation model of the bearing by taking the time sequence training model based on random convolution kernel transformation as a reference, and predicting the characterization parameters of the health condition of the bearing by using the degradation trend evaluation model. Meanwhile, data given by a bearing data set of the university of Sian traffic serve as an experiment and verification object, the established model is substituted to predict the degradation stage of the bearing, and the application range and the effectiveness of the model are finally analyzed. Compared with the traditional method for evaluating the health condition and the degradation trend of the bearing by only selecting one-dimensional time domain feature vectors, the method provides a two-stage fusion method of time-frequency domain features and modal feature parameters, and modal features are prevented from being submerged in a plurality of time-frequency domain features. . A random convolution kernel transform model is used which uses a large number of convolution kernels, benefiting from the fact that learning weights need hardly be considered, and therefore the cost of convolution computation is relatively low, so a large number of kernels can be used with relatively little computational cost.
The related technical scheme is the same as embodiment 1, and is not described herein.
Example 3
A degradation trend evaluation system of a variable condition bearing comprises: the evaluation method comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to execute the evaluation method for the degradation trend of the variable condition bearing provided by the embodiment 1 of the invention.
Example 4
A machine-readable storage medium having stored thereon machine-executable instructions which, when invoked and executed by a processor, cause the processor to perform a method of constructing a variable regime bearing fault diagnosis model as provided by embodiment 1 of the invention and/or a method of diagnosing a variable regime bearing fault as provided by embodiment 2 of the invention.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. A degradation trend evaluation method of a variable condition bearing is characterized by comprising the following steps:
s1, constructing a bearing degradation trend evaluation model; the bearing degradation trend evaluation model comprises a feature extraction module, a feature fusion module, a random convolution kernel transformation module, a dimension reduction module and a label classifier; the characteristic extraction module is used for extracting modal characteristics and time-frequency characteristics of the input data; the characteristic fusion module is used for carrying out characteristic fusion on the time-frequency characteristic and the modal characteristic through typical correlation analysis; the random convolution kernel transformation module is used for performing convolution operation on the fused characteristic parameters through a large number of multi-scale convolution kernels and extracting multi-scale characteristics related to bearing faults and unrelated to the bearing faults; the dimension reduction module is used for carrying out dimension reduction projection on the multi-scale features and constructing health state features; the label classifier is used for mapping the health state features after dimension reduction into corresponding predicted bearing health states;
s2, using vibration signal data of the existing multi-working-condition bearing with the label as source domain data, and using vibration signal data of the bearing without the label in the new working condition as target domain data to form a training sample set;
s3, carrying out iterative training on the bearing degradation trend evaluation model by adopting a training sample set; in the training process, initializing the joint probability distribution of the random convolution kernels to be consistent with the data distribution of a bearing training set; the length of the random convolution kernel is set to be the product of a numerical value of sampling points of more than one circle of the rolling bearing and normal distribution
And S4, inputting vibration data of the bearing to be tested under the current working condition into the trained model to obtain a bearing degradation trend evaluation result.
2. The method for evaluating the degradation trend of the variable condition bearing according to claim 1, wherein the fusion process of the modal characteristics and the time-frequency characteristics comprises the following steps:
analyzing and fusing time-frequency characteristics through typical correlation;
and multiplying the structural modal characteristics by the corresponding coefficients, and performing concat secondary fusion on the time-frequency characteristics subjected to typical correlation analysis fusion.
3. The method for evaluating the degradation tendency of the variable working condition bearing as claimed in claim 2, wherein the modal characteristics are obtained by a covariance-based random subspace identification method SSICOV; the modal characteristics include modal frequency, modal shape and modal damping.
4. The method for evaluating the degradation trend of the variable-condition bearing according to claim 3, wherein the modal characteristics are further preprocessed as follows: taking the first-order natural frequency from the multi-order natural frequency parameters as an effective value; and meanwhile, interpolation is adopted to process missing values.
5. The method for evaluating the degradation trend of a variable-condition bearing according to any one of claims 1-4, wherein the time-domain characteristics are root mean square values and kurtosis indexes.
6. The method for evaluating the degradation tendency of a variable-condition bearing according to claim 5, wherein the frequency domain characteristic is a root mean square frequency.
7. A degradation tendency evaluation device of a variable condition bearing is characterized by comprising:
the evaluation model building module is used for building a bearing degradation trend evaluation model; the bearing degradation trend evaluation model comprises a feature extraction module, a feature fusion module, a random convolution kernel transformation module, a dimension reduction module and a label classifier; the characteristic extraction module is used for extracting modal characteristics and time-frequency characteristics of the input data; the characteristic fusion module is used for carrying out characteristic fusion on the time-frequency characteristic and the modal characteristic through typical correlation analysis; the random convolution kernel transformation module is used for performing convolution operation on the fused characteristic parameters through a large number of multi-scale convolution kernels and extracting multi-scale characteristics related to bearing faults and unrelated to the bearing faults; the dimension reduction module is used for carrying out dimension reduction projection on the multi-scale features and constructing health state features; the label classifier is used for mapping the health state features subjected to dimension reduction into corresponding predicted bearing health states;
the training sample construction module is used for taking vibration signal data of an existing multi-working-condition bearing with a label as source domain data and taking vibration signal data of a bearing without a label under a new working condition as target domain data to form a training sample set;
the iterative training module is used for performing iterative training on the bearing degradation trend evaluation model by adopting a training sample set; in the training process, initializing the joint probability distribution of the random convolution kernels to be consistent with the data distribution of a bearing training set; the length of the random convolution kernel is set to be the product of a numerical value of sampling points of more than one circle of the rolling bearing and normal distribution
And the online evaluation module is used for inputting the vibration data of the bearing to be tested under the current working condition into the trained model to obtain the evaluation result of the degradation trend of the bearing.
8. A computer-readable storage medium, comprising a stored computer program, wherein when the computer program is executed by a processor, the computer program controls a device on which the storage medium is located to execute the degradation tendency assessment method for a variable-regime bearing according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211493210.XA CN115878986A (en) | 2022-11-25 | 2022-11-25 | Degradation trend evaluation method for variable working condition bearing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211493210.XA CN115878986A (en) | 2022-11-25 | 2022-11-25 | Degradation trend evaluation method for variable working condition bearing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115878986A true CN115878986A (en) | 2023-03-31 |
Family
ID=85764081
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211493210.XA Pending CN115878986A (en) | 2022-11-25 | 2022-11-25 | Degradation trend evaluation method for variable working condition bearing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115878986A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116434372A (en) * | 2023-06-12 | 2023-07-14 | 昆明理工大学 | Intelligent data acquisition system and working condition identification system for variable working condition equipment |
-
2022
- 2022-11-25 CN CN202211493210.XA patent/CN115878986A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116434372A (en) * | 2023-06-12 | 2023-07-14 | 昆明理工大学 | Intelligent data acquisition system and working condition identification system for variable working condition equipment |
| CN116434372B (en) * | 2023-06-12 | 2023-08-18 | 昆明理工大学 | Intelligent data acquisition system and working condition identification system for variable working condition equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| RU2522037C2 (en) | Aircraft engine fault identification | |
| Li et al. | Multi-sensor data-driven remaining useful life prediction of semi-observable systems | |
| CN112633339A (en) | Bearing fault intelligent diagnosis method, bearing fault intelligent diagnosis system, computer equipment and medium | |
| Ayodeji et al. | Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN109623489B (en) | Improved machine tool health state evaluation method and numerical control machine tool | |
| Zhang et al. | Remaining Useful Life Prediction of Rolling Bearings Using Electrostatic Monitoring Based on Two‐Stage Information Fusion Stochastic Filtering | |
| CN110555235A (en) | Structure local defect detection method based on vector autoregressive model | |
| CN115878986A (en) | Degradation trend evaluation method for variable working condition bearing | |
| CN114139589A (en) | Fault diagnosis method, device, equipment and computer readable storage medium | |
| Ghorbani et al. | Estimating remaining useful life of turbofan engine using data-level fusion and feature-level fusion | |
| CN110956331A (en) | Method, system and device for predicting operation state of digital factory | |
| Wang et al. | Two-stage imbalanced learning-based quality prediction method for wheel hub assembly | |
| Esfahani et al. | Remaining useful life prognostics based on stochastic degradation modeling: turbofan engine as case study | |
| CN111474476A (en) | Motor fault prediction method | |
| US20220044151A1 (en) | Apparatus and method for electronic determination of system data integrity | |
| CN113127342B (en) | Defect prediction method and device based on power grid information system feature selection | |
| Shu et al. | A Comprehensive working condition identification scheme for rolling bearings based on modified CEEMDAN as well as modified hierarchical amplitude-aware permutation entropy | |
| Brandao et al. | Fault Diagnosis of Rotary Machines Using Machine Learning | |
| Li et al. | Fuzzy feature fusion and multimodal degradation prognosis for mechanical components | |
| CN114184367B (en) | Rotary mechanical equipment fault diagnosis method, device, equipment and readable storage medium | |
| CN115658371B (en) | Diagnosis algorithm quantitative recommendation method based on case learning and diagnosability analysis | |
| CN111507639B (en) | Financing risk analysis method and device | |
| Liu et al. | Refined Composite Multiscale Range Entropy and Pairwise Feature Proximity-Based Fault Detection Method of Rotating Machinery | |
| CN117609737B (en) | Method, system, equipment and medium for predicting health state of inertial navigation system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |